<s>
Best	B-Algorithm
node	I-Algorithm
search	I-Algorithm
(	O
BNS	O
)	O
,	O
originally	O
known	O
as	O
fuzzified	O
game	O
tree	O
search	O
,	O
is	O
a	O
minimax	B-Algorithm
search	B-Application
algorithm	I-Application
,	O
developed	O
in	O
2011	O
.	O
</s>
<s>
The	O
idea	O
is	O
that	O
the	O
knowledge	O
that	O
one	O
subtree	B-Application
is	O
relatively	O
better	O
than	O
some	O
(	O
or	O
all	O
)	O
other(s )	O
may	O
be	O
propagated	O
sooner	O
than	O
the	O
absolute	O
value	O
of	O
minimax	B-Algorithm
for	O
that	O
subtree	B-Application
.	O
</s>
<s>
First	O
an	O
initial	O
guess	O
at	O
the	O
minimax	B-Algorithm
value	O
must	O
be	O
made	O
,	O
possibly	O
based	O
on	O
statistical	O
information	O
obtained	O
elsewhere	O
.	O
</s>
<s>
Then	O
BNS	O
calls	O
search	O
that	O
tells	O
whether	O
the	O
minimax	B-Algorithm
of	O
the	O
subtree	B-Application
is	O
smaller	O
or	O
bigger	O
than	O
the	O
guess	O
.	O
</s>
<s>
It	O
changes	O
the	O
guessed	O
value	O
until	O
alpha	B-Language
and	O
beta	B-Language
are	O
close	O
enough	O
or	O
only	O
one	O
subtree	B-Application
allows	O
a	O
minimax	B-Algorithm
value	O
greater	O
than	O
the	O
current	O
guess	O
.	O
</s>
<s>
The	O
search	O
result	O
is	O
the	O
node	O
(	O
move	O
)	O
whose	O
subtree	B-Application
contains	O
the	O
minimax	B-Algorithm
value	O
,	O
and	O
a	O
bound	O
on	O
that	O
value	O
,	O
but	O
not	O
the	O
minimax	B-Algorithm
value	O
itself	O
.	O
</s>
<s>
Experiments	O
with	O
random	O
trees	O
show	O
it	O
to	O
be	O
the	O
most	O
efficient	O
minimax	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Tree	B-Application
searching	I-Application
with	O
Murphy	O
Sampling	O
is	O
an	O
extension	O
of	O
Best	B-Algorithm
Node	I-Algorithm
Search	I-Algorithm
to	O
non-deterministic	O
setting	O
.	O
</s>
